Playwright SEO Audit

Checks the mechanical, verifiable layer of SEO — does a title tag exist and is it a sane length, does a page meant to be indexed accidentally say otherwise, does structured data actually parse — not the strategic layer, which isn't something a browser session can verify.

Scope boundary

This stays in checkable territory: presence, validity, structural correctness, uniqueness. It does not predict ranking impact, judge keyword choices, or give content strategy advice — those require actual search engine behavior or editorial judgment this skill has no way to verify, the same reason a general "improve the visual design" request was ruled out of this toolkit earlier. If asked to rank pages by SEO quality or predict search performance, redirect back to what's actually checkable instead.

Relationship to the other skills

Reuse scenario-mapper's page list. Don't duplicate accessibility-audit's territory — image alt text is already that skill's job (a literal axe-core rule), even though it has SEO value too; cross-reference rather than re-checking it here. Reuse link-audit's lightweight-HTTP-check technique for verifying sitemap URLs actually resolve, rather than re-deriving it. For a multi-locale site, hreflang tag correctness (whether locale variants correctly reference each other) is i18n-audit's job, not this one's — this skill validates a single page's metadata in isolation.

Core principles (and why)

An accidental noindex or a Disallow: / in robots.txt is the single highest-value thing this check can catch. Both are extremely common — usually leftover from a staging config that never got cleaned up — and extremely damaging, since either can make an entire site or section invisible to search engines while everything otherwise looks fine. Prioritize catching these over subtler issues like description length.

Structured data validation here is a first-pass check, not full schema compliance. Confirm each JSON-LD block is syntactically valid JSON, note its declared @type, and flag if commonly-required fields for that type are obviously missing — but say plainly that full schema.org conformance needs a dedicated validator (Google's Rich Results Test is the authoritative one) if the site actually depends on rich results. Don't imply deeper validation than what was actually done.

Duplicate titles/descriptions across pages are one finding, not N findings. If the same title appears on twelve pages, that's "not unique, affects 12 pages" — one row, same aggregation instinct as accessibility-audit's "group by rule, note affected-page count."